Hybrid brain-computer interface using error-related potential and reinforcement learning.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Non-invasive brain-computer interfaces (BCIs) using electroencephalography (EEG) face performance limitations due to signal non-stationarities.
• Adaptive systems are crucial for real-time adjustment in BCIs to overcome these limitations.

Purpose of the Study:

• To develop an adaptive, error-related potential (ErrP)-based BCI system utilizing reinforcement learning (RL).
• To dynamically adapt BCIs to electroencephalography (EEG) signal variations in real-time.

Main Methods:

• Implemented a novel adaptive BCI framework employing reinforcement learning (RL).
• Validated the system using a public motor imagery dataset and a custom fast-paced protocol.
• Trained RL agents to learn control policies from user interactions and adapt to EEG signal changes.

Main Results:

• Reinforcement learning agents successfully learned control policies and maintained robust performance across different datasets.
• The study identified that fast-paced motor imagery tasks in a game-based protocol were largely ineffective for participants.
• Demonstrated the potential of RL for enhancing BCI adaptability.

Conclusions:

• Reinforcement learning shows promise for improving the adaptability of brain-computer interfaces (BCIs).
• Practical challenges exist in designing real-time BCI tasks, particularly concerning task complexity and user responsiveness.
• Further research is needed to optimize BCI task design for effective user engagement and performance.